1. Introduction and problem

In the 3-to-6 age band —kindergarten, preschool, CENDI, early childhood school— a package of five artefacts has been installed as if it counted as support for risky play: a wearable or sensor that “scores fall risk” or a “fall-risk score”; a computer-vision / playground-camera system that labels “dangerous play” or triggers “unsafe play” alerts; a chatbot/GenAI that generates “safe alternative activities” or “risk-free play ideas” by prompt; a dashboard of “risk metrics” / “risk exposure score” for administrative surveillance; and an AR layer that “guides” the child to avoid height or speed. The five allow centres to exhibit that they “already do risky play with AI.” The leap —from biometric score, unsafe-play alert, “safe” alternative, exposure metric or AR guide to claiming support— is not authorized by AI-in-ECE mappings nor by the pedagogy of risky play when there are real physical opportunities (height, speed, tools, natural elements, rough-and-tumble, disappearing/being alone), thrill, outcome uncertainty, possibility of minor injury, child agency to assess and handle risk, and adult co-presence that distinguishes risk from hazard.

This article’s thesis is restrictive. A wearable or sensor that “scores fall risk” or a “fall-risk score,” a computer-vision / playground-camera system that labels “dangerous play” or triggers “unsafe play” alerts, a chatbot/GenAI that generates “safe alternative activities” or “risk-free play ideas” by prompt, a dashboard of “risk metrics” / “risk exposure score” for administrative surveillance, or an AR layer that “guides” the child to avoid height/speed do not constitute support for risky play in the early years. In ECE, risky play develops with real physical opportunities, child agency to assess and handle risk, adult co-presence that distinguishes risk from hazard, and calibrated adult tolerance (e.g., OutsidePlay-ECE as human professional development); not a biometric score, an unsafe-play classifier, a generator of “safe” alternatives that removes uncertainty, or a surveillance dashboard. Sando, Kleppe and Sandseter (2021) document, with N = 928 video observations in eight Norwegian ECEC centres, associations between risky play and well-being, involvement and physical activity: an empirical finding of situated practice, not of a fall-risk score. Sandseter, Kleppe and Kennair (2023) articulate an evolutionary biopsychosocial framework on emotion regulation, social functioning and physical health. Kleppe, Sandseter, Sando and Brussoni (2024) conceptualize children’s dynamic risk management —willingness, assessment, handling. Beaulieu and Beno (2024), from Paediatrics & Child Health / CPS, fix the risk ≠ hazard distinction and the balance with injury prevention. That authorizes asking what was measured: fall score, unsafe-play label, “safe” alternative by prompt, risk exposure score or AR guide —not the practice when a child calibrates height or speed, assesses and handles uncertainty, and receives adult co-presence that distinguishes hazard from opportunity.

The problem is worsened by five category confusions. First: risky play is not generic outdoor / nature play —although it may occur outdoors, the axis is the child’s risk calibration (Sandseter categories), not outdoor minutes or green time already published. Second: it is not generic “free play” —free play may host it; the axis here is thrill + uncertainty + possibility of minor injury with dynamic management. Third: it is not curricular motor development as axis —physical activity may be a lateral outcome (Sando et al., 2021). Fourth: it is not SEL as axis —Sandseter et al. (2023) discuss emotion regulation in an evolutionary key of risky play, not a published SEL curriculum. Fifth: it is not a fall-risk score, unsafe-play alert, risk-free alternative or surveillance dashboard. This work does not recycle outdoor play, free play, loose parts, block play/spatial reasoning, symbolic play, motor development, STEM/robotics, artistic creativity, SEL, executive functions, environmental education, inquiry, literacy, orality, UDL, documentation, formative assessment, participation, planning, joint attention, adult–child interactions as a generic axis, teacher education, music or numeracy. Brussoni et al. (2022) —OutsidePlay-ECE— is used only as a contrast of peripheral good use (EDUCATOR formation / human reframing), not as an outdoor axis nor as already-published continuing education. The question is what counts as support for risky play when a centre “does AI and risky play.”

There is also a coordination economy: fall-risk wearable, unsafe-play camera, GenAI risk-free alternatives, risk-exposure dashboard and AR layer fit on a slide; an episode of climbing, running, using tools, rough-and-tumble or disappearing/being alone with child agency and co-presence that calibrates hazard does not. NAEYC (2022) and OECD (2021, 2023) require meaningful interactions and subordinated digitalization. Inference: support for risky play is not fulfilled by scoring falls or generating “risk-free activities.” The contributions are three: separating craft from the artefact package; examining three case families —including OutsidePlay-ECE (Brussoni et al., 2022) as a peripheral contrast toward the adult and Liu and Birkeland (2022) as cultural contrast—; and offering four tests to affirm support for risky play, not merely wearable, CV, GenAI, dashboard or AR.

2. State of the art: from situated risky-play practice to the exhibited artefact

Four strata that the “AI for risky play in ECE” market usually mixes should be separated. The first is the construct of risky play for ages 3–6 as practice with Sandseter categories —height, speed, tools, natural elements, rough-and-tumble, disappearing/being alone—, thrill, outcome uncertainty, possibility of minor injury, and children’s dynamic risk management (willingness, assessment, handling) (Sando et al., 2021; Sandseter et al., 2023; Kleppe et al., 2024; Beaulieu & Beno, 2024; MacQuarrie et al., 2022). The second is the pedagogical craft that cultivates it —provision of real physical opportunities, adult co-presence that distinguishes risk from hazard, calibrated tolerance, not substituting the child’s calibration with a score or an unsafe-play alert— (Kleppe et al., 2024; Beaulieu & Beno, 2024; Liu & Birkeland, 2022; Brussoni et al., 2022, only as a human/web tool for the educator; NAEYC, 2022; OECD, 2021, 2023). The third is evidence on AI affordances in ECE and child-centred GenAI —without equating them to situated risky play with body, uncertainty and agency— (Chen, 2024; Su & Yang, 2022; Su & Zhong, 2022; Ljungcrantz, 2026; Nikolopoulou, 2025). The fourth is the rights, systems and developmentally appropriate practice framework that treats the 3–6-year-old as a subject who calibrates risk, not as a signal vector for a fall-risk pipeline nor as a passive recipient of “safe” alternatives that remove uncertainty (UNESCO, 2021; Miao & Holmes, 2023; U.S. Department of Education, 2023).

In construct and craft, Sando et al. (2021) —N = 928 video observations; eight Norwegian ECEC centres— associate risky play with well-being, involvement and physical activity: empirical finding. Sandseter et al. (2023) contribute an evolutionary biopsychosocial framework. Kleppe et al. (2024) articulate willingness, assessment and handling: craft framework. Beaulieu and Beno (2024) fix risk ≠ hazard from Canadian paediatrics. MacQuarrie et al. (2022) explore parental perceptions with a socio-ecological lens; marked transfer (not a wearable trial). Liu and Birkeland (2022) —N = 10; Norway–Anji— contrast cultural perceptions. Inference: a fall-risk score does not calibrate thrill or uncertainty; a child who assesses height and an adult who distinguishes hazard from opportunity do.

In the artefact stratum, Chen (2024), Su and Yang (2022), Su and Zhong (2022) and Ljungcrantz (2026) map affordances, curricula and the state of the art of AI in ECE without equating them to situated risky play. Nikolopoulou (2025) balances promises and challenges of child-centred GenAI under teacher mediation: caution framework. Brussoni et al. (2022) evaluate OutsidePlay-ECE in a Canadian RCT: a web intervention that influences educators’ attitudes and supportive behaviours toward outdoor play and risk tolerance (effects at 1 week and 3 months): empirical finding of a human/web tool for the EDUCATOR —legitimate peripheral contrast—, not an AI score on the child nor an unsafe-play classifier. In the systems stratum, UNESCO (2021) requires human oversight. Miao and Holmes (2023) set pedagogical validation and age thresholds for generative AI —direct framework for the generator of “safe activities” by prompt—. U.S. Department of Education (2023) requires that AI support, not replace, professional judgement. OECD (2021, 2023) anchor meaningful interactions and subordinated digitalization; NAEYC (2022) anchors developmentally appropriate practice.

3. Review method

A critical narrative review was conducted, not a primary meta-analysis. The purpose was not to estimate a homogeneous effect size of the five artefacts, but to articulate a pedagogical-category argument with verified sources. Inclusion criteria: (a) 2021–2026, with marked transfer when the sample is not equivalent to ages 3–6; (b) risky play, Sandseter categories, dynamic risk management, risk ≠ hazard, or AI in ECE with artefact/craft relevance; (c) kindergarten, preschool, CENDI or ages 3–6; (d) peer-reviewed journal, DOI or NAEYC/UNESCO/OECD report; (e) verifiable DOI or editorial page. Excluded as central object were axes already used in this series —generic outdoor play, free play, loose parts, block play/spatial, symbolic play, STEM/robotics, artistic creativity, motor development as axis, SEL, executive functions, environmental education, inquiry, literacy, orality, UDL, documentation, formative assessment, participation, planning, joint attention, adult–child interactions as generic axis, teacher education, music and numeracy—. OutsidePlay-ECE (Brussoni et al., 2022) only as a peripheral good-use contrast (tool→adult/educator); the article is not converted into outdoor play nor into already-published continuing education. Outdoor risky play in Beaulieu and Beno (2024) and MacQuarrie et al. (2022) is read for risk/hazard calibration and perceptions, not as an outdoor-minutes axis.

The search was executed on 3 September 2026 (slot 05:02 America/Mexico_City) on DOI pages, Crossref, Springer, Elsevier, Taylor & Francis, JMIR, Oxford Academic, JAIR, OECD iLibrary, UNESDOC, NAEYC and editorial sites. Each source was verified. Empirical finding, framework and pedagogical inference were distinguished. No N, d, r, AUC or DOI was invented: when an artefact lacks a verified study with fall-risk scoring, unsafe-play alerts, risk-exposure dashboard or anti-height/speed AR metrics in ages 3–6, it is discussed as a category ceiling supported by AI-in-ECE mappings (Chen, 2024; Su & Yang, 2022; Ljungcrantz, 2026) and by the OutsidePlay-ECE contrast (Brussoni et al., 2022). The eighteen sources of the verified corpus were used.

4. Case 1. Wearable/fall-risk score, CV/playground camera with unsafe-play alerts, GenAI “safe activities” / risk-free alternatives, risk-metrics dashboard or AR that avoids height/speed do not constitute support for risky play

Chen (2024), Su and Yang (2022) and Ljungcrantz (2026) saturate the portrait of the artefact ceiling when AI in ECE is presented as if it were support for risky play. Chen (2024) maps global AI affordances in early childhood education: a scoping finding on emerging uses —tutoring, analytics, content generation—, not a finding that a fall-risk wearable cultivates willingness, assessment and handling (Kleppe et al., 2024). Su and Yang (2022) review the AI-in-ECE field: a synthesis finding on trends, not on thrill, outcome uncertainty or mediated rough-and-tumble. Ljungcrantz (2026) reviews AI–ECE interaction 2020–2024: a state-of-the-art finding, not of situated risky play. Pedagogical inference, marked as such: the gesture “the child had a low fall-risk score = there was support for risky play” is a biometric-analytics ceiling. Sando et al. (2021), Sandseter et al. (2023) and Kleppe et al. (2024) require real physical opportunities, dynamic management by the child and co-presence that calibrates; a score useful for administration can coexist with absence of height, speed, tools, rough-and-tumble and calibrated adult tolerance.

Nikolopoulou (2025) and Miao and Holmes (2023) name the risk of the generator of “safe alternative activities” or “risk-free play ideas” from a prompt that removes uncertainty. Nikolopoulou (2025) balances promises and challenges of child-centred GenAI under teacher mediation: caution framework. Miao and Holmes (2023) require pedagogical validation and age thresholds for generative AI. Status: normative and review framework, not a trial of risk-free alternatives by prompt versus a risky-play episode with Sandseter categories and co-presence that distinguishes risk from hazard (Beaulieu & Beno, 2024). Inference: producing a list of “safe activities” by prompt may be subordinated teacher preparation; support for risky play begins when there is a real physical opportunity, child agency to assess and handle, thrill and uncertainty, and an adult who calibrates hazard versus opportunity without eliminating risk. Su and Zhong (2022) propose AI curriculum design in ECE as a future direction —AI literacy curriculum, not situated risky play—. Limit inference: an AI curriculum does not sign the child’s risk calibration.

The computer-vision / playground-camera system that labels “dangerous play” or triggers “unsafe play” alerts, the dashboard of “risk metrics” / “risk exposure score” and the AR layer that “guides” the child to avoid height/speed lack, in the verified corpus, trials with reported N, d, r or AUC for fall-risk scoring, unsafe-play classification, risk-exposure dashboard or anti-height AR guidance in preschool 3–6; figures are not invented. Chen (2024), Su and Yang (2022) and Ljungcrantz (2026) map analytics, tutoring, agents and generation as an AI-in-ECE trend, without equivalence to mediated risky play. Sando et al. (2021) measure human video observations (N = 928): a finding of associations with well-being, involvement and physical activity, not of an autonomous unsafe-play classifier. Brussoni et al. (2022) —OutsidePlay-ECE RCT— document change in educators’ supportive attitudes and behaviours: empirical contrast of legitimate peripheral use only when the tool supports the adult/educator. Restrictive inference: OutsidePlay-ECE (tool→educator) ≠ fall-risk wearable on the child; GenAI of “safe” alternatives ≠ calibrated tolerance that preserves uncertainty; teacher empowerment ≠ administrative risk-exposure score; AR layer that avoids height/speed ≠ Sandseter height/speed category as opportunity. UNESCO (2021) and U.S. Department of Education (2023) require human oversight. The five artefacts share the same substitution grammar: the fall-risk score speaks for the child’s assessment; the unsafe-play alert replaces the risk≠hazard judgement; GenAI removes uncertainty; the dashboard replaces pedagogical observation; AR directs avoidance of what the craft offers as calibration.

The ceiling should be specified without inventing effects. A wearable may score “fall risk”; Kleppe et al. (2024) articulate human willingness–assessment–handling, not biometric AUC. A camera may label “unsafe play”; Beaulieu and Beno (2024) require distinguishing risk from hazard. A GenAI may print “risk-free activities”; Miao and Holmes (2023) require pedagogical validation. An AR may avoid height; Sando et al. (2021) associate risky play with well-being and involvement. A dashboard may raise “risk exposure”; NAEYC (2022) and OECD (2021) anchor practice. Inference: “low fall-risk” or “safe alternative per the model” is measurement or generation that removes uncertainty, not signed support.

5. Case 2. What the kindergarten does when there is support for risky play: Sandseter categories, risk ≠ hazard, dynamic management —with OutsidePlay-ECE (Brussoni) and Liu & Birkeland as contrast

Sando, Kleppe and Sandseter (2021) saturate the empirical floor: N = 928 video observations in eight Norwegian ECEC centres associate risky play with well-being, involvement and physical activity. Status: empirical finding. Marked transfer: the context is Norwegian ECEC; it is read with caution toward Latin American CENDI/kindergarten, not as a fall-risk-score trial. Pedagogical inference, marked as such: this is the kind of object an ECE centre may call support for risky play when it protects real physical opportunities and child agency —not when it scores fall risk on a wearable—. Risky play is not a fall-risk score: it is growing mastery of climbing (height), running/sliding (speed), using tools, playing with natural elements, rough-and-tumble, disappearing/being alone, with thrill, outcome uncertainty and possibility of minor injury, while the child exercises willingness, assessment and handling (Kleppe et al., 2024) and the adult calibrates hazard versus opportunity (Beaulieu & Beno, 2024). A wearable that “protects” without allowing calibration may celebrate a low score and, at once, empty agency. A camera may label “unsafe play” and not have captured whether there was real hazard, whether the adult distinguished risk from danger, or whether the child was measuring their own willingness.

Sandseter et al. (2023) contribute the evolutionary biopsychosocial framework: risky play in emotion regulation, social functioning and physical health. Status: conceptual framework. Kleppe et al. (2024) articulate dynamic risk management —willingness, assessment, handling— as a comprehensive approach: the child is not a passive recipient of algorithmic “safety”; they are an agent who assesses and handles. Beaulieu and Beno (2024) fix, from Canadian paediatrics (CPS / Paediatrics & Child Health), that healthy development passes through outdoor risky play navigating the balance with injury prevention: risk ≠ hazard. MacQuarrie et al. (2022) explore parental perceptions of preschoolers’ risky outdoor play with a socio-ecological lens: the craft is also read from families; marked transfer (parental perceptions ≠ dashboard trial). NAEYC (2022) and OECD (2021, 2023) anchor meaningful interactions and subordinated digitalization.

The peripheral good-use contrast is OutsidePlay-ECE (Brussoni et al., 2022). They report a Canadian RCT of a web intervention for early childhood educators: OutsidePlay-ECE influences attitudes and supportive behaviours toward outdoor play and risk tolerance, with effects at one week and three months. Status: empirical finding of a human/web tool oriented to the EDUCATOR. Inference: subordinated digital —an adult who receives web formation/reframing to raise calibrated tolerance without substituting the child’s agency or calibration— can be coherent with risky play; it does not authorize declaring that a fall-risk wearable, an unsafe-play camera, a GenAI of “safe” alternatives, a risk-exposure dashboard or an anti-height/speed AR constitute the craft. OutsidePlay-ECE is not converted here into a generic outdoor-play article nor into already-published continuing education: it is cited only as a peripheral good-use contrast (tool→adult). Liu and Birkeland (2022) —N = 10 kindergarten teachers; Norway and China (Anji)— contrast cultural perceptions of risky play: the craft is not a universal score; it is a situated reading of tolerance and opportunity. Mediation —NAEYC (2022), OECD (2021); Beaulieu & Beno (2024); Kleppe et al. (2024)— is adult co-presence that organizes physical opportunities, distinguishes risk from hazard, tolerates thrill and uncertainty, observes the child’s willingness–assessment–handling and expands without imposing a digital script that eliminates risk. Inference: pedagogical observation of risky play is a professional reading of the calibration episode; the dashboard counts a surveillance metric. A dashboard does not assess or handle risk; a child and an adult who distinguishes hazard from opportunity do.

6. Case 3. Risky play ≠ generic outdoor/nature play, free play, curricular motor development or SEL as axis

The first category frontier is already-published generic outdoor / nature play. Risky play may occur outdoors; Beaulieu and Beno (2024) and MacQuarrie et al. (2022) speak of outdoor risky play —but this article’s axis is risk calibration (Sandseter categories, thrill, uncertainty, dynamic management), not outdoor minutes, green time, forest school or nature play as a time-outdoors construct. Inference: this article forbids the equivalence “we have a yard / outdoor = there is support for risky play.” OutsidePlay-ECE (Brussoni et al., 2022) is cited for its effect on adult tolerance, not as identity with already-treated outdoor play. A fall-risk wearable is not outdoor either: it is biometric analytics. The second frontier is generic free play: free play may host risky play; it does not, by itself, constitute the practice of thrill + uncertainty + possibility of minor injury with willingness–assessment–handling. A GenAI of “safe activities” is not free either: it removes the uncertainty that defines the construct. The third is curricular motor development as axis: Sando et al. (2021) report physical activity as an associated outcome; they do not authorize reducing risky play to a motor curriculum or to fall tracking. The fourth is SEL as axis: Sandseter et al. (2023) articulate emotion regulation in an evolutionary key of risky play —lateral framework—, not an already-published SEL curriculum. The fifth is the fall-risk score / unsafe-play alert / risk-exposure dashboard / anti-height AR: scoring, alerting, surveilling or guiding avoidance inverts the pedagogy of calibration. UNESCO (2021), Miao and Holmes (2023) and U.S. Department of Education (2023) require human oversight. Inference: the only AI use coherent with ages 3–6 remains on the adult side —OutsidePlay-ECE as peripheral contrast (Brussoni et al., 2022)—, subject to pedagogical validation and contrasted with the child’s dynamic management (Kleppe et al., 2024) and risk ≠ hazard (Beaulieu & Beno, 2024). It does not enter as biometric scorer, unsafe-play classifier, risk-free-alternative generator, surveillance dashboard or AR that avoids height/speed.

7. Inferential framework: four tests to affirm that there is support for risky play, not an artefact

The following framework is pedagogical inference of this article, anchored in the cases and verified instruments. It is not a new international standard. It distinguishes four tests. If a kindergarten, preschool, CENDI or early childhood school does not pass them, it cannot declare that the five artefacts constitute support for risky play.

7.1. Test of situated practice of real physical opportunities (Sandseter categories), thrill, uncertainty and child agency, not of the fall-risk wearable nor of the anti-height/speed AR layer. Sando et al. (2021), Sandseter et al. (2023) and Kleppe et al. (2024) define the craft as observable risky play with dynamic management. Chen (2024), Su and Yang (2022) and Ljungcrantz (2026) map analytics and generation as AI affordances, not as risky play. Inference: evidence of support for risky play is verified in whether the child had a real physical opportunity (height, speed, tools, natural elements, rough-and-tumble, disappearing/being alone) with agency to assess and handle. If the centre’s “evidence” is a fall-risk score or an AR guide that avoids height/speed, the centre has done analytics or digital anti-calibration direction, not support for risky play.

7.2. Test of the risk ≠ hazard distinction and of adult co-presence that calibrates, not of GenAI “safe activities” / risk-free alternatives. Beaulieu and Beno (2024), MacQuarrie et al. (2022) and NAEYC (2022) situate risk/hazard, perceptions and developmentally appropriate practice. Miao and Holmes (2023) require pedagogical validation of generative AI. Inference: producing “safe” alternatives by prompt that remove uncertainty does not demonstrate that there was thrill, possibility of minor injury or an adult who distinguishes hazard from opportunity. A risk-free script may exist; it does not sign the risky-play episode.

7.3. Test of human mediation —including the OutsidePlay-ECE peripheral contrast—, not of the risk-exposure dashboard nor of the CV/camera unsafe-play alerts. OECD (2021, 2023) require meaningful interactions and subordinated digitalization. Brussoni et al. (2022) anchor web tool→educator with tolerance reframing. Kleppe et al. (2024) anchor the child’s willingness–assessment–handling. Inference: “low risk exposure score” or “unsafe-play alert” may raise an administrative threshold without raising the quality of co-presence or child calibration. Pedagogical observation reads the dynamic-management episode; the dashboard counts surveillance. OutsidePlay-ECE that forms the adult is legitimate peripheral use; the wearable/CV that substitutes the child’s assessment or the risk≠hazard distinction is not.

7.4. Test of category distinction and professional judgement, not of the product catalogue. Risky play ≠ generic outdoor/nature play, free play, curricular motor development or SEL as axis. Sando et al. (2021) prevent reducing associations with physical activity to a motor curriculum. Sandseter et al. (2023) prevent confusing evolutionary emotion regulation with curricular SEL. Liu and Birkeland (2022) prevent treating the construct as a culturally blind score. UNESCO (2021), Miao and Holmes (2023), U.S. Department of Education (2023), NAEYC (2022) and OECD (2021, 2023) require human oversight, pedagogical validation and not replacing professional judgement. Inference: a centre cannot treat the infant as a fall-risk vector nor as the exclusive addressee of risk-free alternatives. Support for risky play is not fulfilled by better algorithmic “safety” scoring. It is fulfilled by practising real physical opportunities, child agency to assess and handle risk, adult co-presence that distinguishes risk from hazard, and calibrated adult tolerance.

The framework admits subordinated digital for adult formation/reframing and validated human mediation (Brussoni et al., 2022; Chen, 2024; Su & Yang, 2022; Ljungcrantz, 2026). It rejects declaring support via the five artefacts (Miao & Holmes, 2023; Nikolopoulou, 2025; UNESCO, 2021). The four tests are read together.

8. Discussion

Three tensions organize the discussion. The first is between exhibiting wearable, CV unsafe-play, GenAI risk-free, dashboard or AR and exercising support for risky play. It is a finding that N = 928, the evolutionary framework, dynamic management, risk≠hazard, parental perceptions, cultural contrast N = 10 and OutsidePlay-ECE sustain the craft or its frontier (Sando et al., 2021; Sandseter et al., 2023; Kleppe et al., 2024; Beaulieu & Beno, 2024; MacQuarrie et al., 2022; Liu & Birkeland, 2022; Brussoni et al., 2022). It is a framework that AI in ECE grows without equivalence to situated risky play (Chen, 2024; Su & Yang, 2022; Ljungcrantz, 2026). It is not a finding that the five artefacts produce the craft.

The second is between automated assessment of “fall-risk” or “unsafe play” and the pedagogy of calibration. Chen (2024), Su and Yang (2022) and Ljungcrantz (2026) objectivize analytics as a trend; they do not report verified trials of fall-risk scoring, unsafe-play AUC or risk-exposure dashboard in preschool 3–6 —metrics are not invented—. Inference: declaring support because a wearable scores falls or a camera alerts “dangerous play” is inverted pedagogy: the proxy is made to stand for physical opportunities, agency, dynamic management and co-presence.

The third is between GenAI of “safe” alternatives / AR guidance and integral craft. Miao and Holmes (2023) and Nikolopoulou (2025) fix GenAI limits; Brussoni et al. (2022) fix human adult reframing; Kleppe et al. (2024) and Beaulieu and Beno (2024) fix child handling and risk≠hazard. Inference: selling risk-free alternatives or anti-height AR as “risky play with AI” confuses textual product or digital direction with calibration practice. Sando et al. (2021) confirm associations of risky play with well-being and involvement without authorizing a score; UNESCO (2021), Miao and Holmes (2023) and U.S. Department of Education (2023) subordinate AI to professional judgement.

The four tests in section 7 read these tensions with operational criteria. The empirical craft contrast —N = 928; evolutionary framework; dynamic management; CPS risk≠hazard; OutsidePlay-ECE; N = 10 cultural— defines the floor the five artefacts do not reach alone. Inference: support erodes when wearable, CV, GenAI, dashboard or AR are treated as if they were the practice. Subordinated digital is legitimate —OutsidePlay-ECE toward the adult (Brussoni et al., 2022)—; inverting the sequence is not (NAEYC, 2022; OECD, 2021, 2023; Kleppe et al., 2024).

9. Limits

This review is narrative. It does not apply its own PRISMA nor estimate primary combined effects. Sando et al. (2021) measure N = 928 in Norwegian ECEC: marked partial transfer toward other contexts. Sandseter et al. (2023) contribute an evolutionary framework; Kleppe et al. (2024), a dynamic-management framework; Beaulieu and Beno (2024), a CPS paediatric framework. MacQuarrie et al. (2022) contribute parental perceptions; Liu and Birkeland (2022), N = 10 Norway–China teachers; Brussoni et al. (2022), OutsidePlay-ECE RCT as peripheral contrast (tool→educator), not as outdoor axis nor continuing education. Chen (2024), Su and Yang (2022), Su and Zhong (2022), Ljungcrantz (2026) and Nikolopoulou (2025) map AI in ECE, not fall-risk scoring or unsafe-play alerts. No verified trials of fall-risk wearable, CV unsafe-play, GenAI risk-free, risk-exposure dashboard or anti-height/speed AR in CENDI ages 3–6 were located; they are discussed as a category ceiling. NAEYC, UNESCO and OECD are framework sources. Section 7 inferences are category hypotheses, not implementation evidence.

10. Conclusions

A wearable or sensor that “scores fall risk” or a “fall-risk score,” a computer-vision / playground-camera system that labels “dangerous play” or triggers “unsafe play” alerts, a chatbot/GenAI that generates “safe alternative activities” or “risk-free play ideas” by prompt, a dashboard of “risk metrics” / “risk exposure score” for administrative surveillance, or an AR layer that “guides” the child to avoid height/speed do not constitute support for risky play in an early childhood education centre. Chen (2024), Su and Yang (2022) and Ljungcrantz (2026) confirm AI affordances without equivalence to situated risky play. Nikolopoulou (2025) and Miao and Holmes (2023) fix GenAI limits. Brussoni et al. (2022) require reading OutsidePlay-ECE as a peripheral contrast —tool→adult ≠ fall-risk wearable, CV unsafe-play, GenAI risk-free, dashboard or AR—. Kleppe et al. (2024) and Beaulieu and Beno (2024) fix dynamic management and risk≠hazard. When there is support, there is situated practice: N = 928 observations (Sando et al., 2021); evolutionary framework (Sandseter et al., 2023); willingness–assessment–handling (Kleppe et al., 2024); risk≠hazard (Beaulieu & Beno, 2024); parental perceptions (MacQuarrie et al., 2022); cultural contrast (Liu & Birkeland, 2022); DAP and interactions (NAEYC, 2022; OECD, 2021, 2023). Risky play is distinguished from generic outdoor/nature play, free play, curricular motor development and SEL as axis. Guidance requires human oversight (UNESCO, 2021; Miao & Holmes, 2023; U.S. Department of Education, 2023; Su & Zhong, 2022).

Where sources do not measure a kindergarten, this article does not affirm it. Where they measure AI mappings, GenAI or OutsidePlay-ECE toward the educator, it does not translate them into pedagogical risky-play support via score or alert. Accompanying three-to-six-year-olds in risky play is to exercise real physical opportunities (Sandseter categories), thrill and uncertainty, child agency to assess and handle risk, adult co-presence that distinguishes risk from hazard, and calibrated adult tolerance. The rest is fall-risk wearable, unsafe-play classifier, “safe”-alternative generator, surveillance dashboard and AR layer that avoids height/speed. It is not support for risky play in early childhood education, and it must not be presented as what it is not.

Laboratorio Editorial de NEXTECH.IA / Ingeniero Mitre.

References

  1. Beaulieu, É., y Beno, S. (2024). Healthy childhood development through outdoor risky play: Navigating the balance with injury prevention. Paediatrics & Child Health, 29(4), 255–261. https://doi.org/10.1093/pch/pxae016
  2. Brussoni, M., Han, C. S., Lin, Y., Jacob, J., Munday, F., Zeni, M., Walters, M., y Oberle, E. (2022). Evaluation of the web-based OutsidePlay-ECE intervention to influence early childhood educators’ attitudes and supportive behaviors toward outdoor play: Randomized controlled trial. Journal of Medical Internet Research, 24(6), e36826. https://doi.org/10.2196/36826
  3. Chen, J. J. (2024). A scoping study on AI affordances in early childhood education. Journal of Artificial Intelligence Research, 81, 701–740. https://doi.org/10.1613/jair.1.16882
  4. Kleppe, R., Sandseter, E. B. H., Sando, O. J., y Brussoni, M. (2024). Children’s dynamic risk management – a comprehensive approach to children’s risk willingness, risk assessment, and risk handling. International Journal of Play, 13(4), 395–409. https://doi.org/10.1080/21594937.2024.2425539
  5. Liu, J., y Birkeland, Å. (2022). Perceptions of risky play among kindergarten teachers in Norway and China. International Journal of Early Childhood, 54(3), 339–360. https://doi.org/10.1007/s13158-021-00313-8
  6. Ljungcrantz, L. (2026). The interaction of AI and early childhood education. A state-of-the-art review 2020–2024. Early Childhood Education Journal, 54, 3565–3581. https://doi.org/10.1007/s10643-025-02079-3
  7. MacQuarrie, M., McIsaac, J.-L. D., Cawley, J., Kirk, S., Kolen, A. M., Rehman, L., y colaboradores. (2022). Exploring parents’ perceptions of preschoolers’ risky outdoor play using a socio-ecological lens. European Early Childhood Education Research Journal, 30(3), 372–387. https://doi.org/10.1080/1350293X.2022.2055103
  8. Miao, F., y Holmes, W. (2023). Guidance for generative AI in education and research. UNESCO. https://doi.org/10.54675/EWZM9535
  9. NAEYC. (2022). Developmentally appropriate practice in early childhood programs serving children from birth through age 8 (4.ª ed.). NAEYC. https://www.naeyc.org/resources/pubs/books/dap-fourth-edition
  10. Nikolopoulou, K. (2025). Child-centered integration of generative AI in early learning: Balancing promises and challenges. AI, Brain and Child. https://doi.org/10.1007/s44436-025-00023-1
  11. OECD. (2021). Starting Strong VI: Supporting meaningful interactions in early childhood education and care. OECD Publishing. https://doi.org/10.1787/f47a06ae-en
  12. OECD. (2023). Empowering young children in the digital age (Starting Strong). OECD Publishing. https://doi.org/10.1787/50967622-en
  13. Sando, O. J., Kleppe, R., y Sandseter, E. B. H. (2021). Risky play and children’s well-being, involvement and physical activity. Child Indicators Research, 14(4), 1435–1451. https://doi.org/10.1007/s12187-021-09804-5
  14. Sandseter, E. B. H., Kleppe, R., y Kennair, L. E. O. (2023). Risky play in children’s emotion regulation, social functioning, and physical health: An evolutionary approach. International Journal of Play, 12(1), 127–139. https://doi.org/10.1080/21594937.2022.2152531
  15. Su, J., y Yang, W. (2022). Artificial intelligence in early childhood education: A scoping review. Computers and Education: Artificial Intelligence, 3, 100049. https://doi.org/10.1016/j.caeai.2022.100049
  16. Su, J., y Zhong, Y. (2022). Artificial Intelligence (AI) in early childhood education: Curriculum design and future directions. Computers and Education: Artificial Intelligence, 3, 100072. https://doi.org/10.1016/j.caeai.2022.100072
  17. U.S. Department of Education, Office of Educational Technology. (2023). Artificial intelligence and the future of teaching and learning: Insights and recommendations. U.S. Department of Education. https://www.ed.gov/sites/ed/files/documents/ai-report/ai-report.pdf
  18. UNESCO. (2021). Recommendation on the ethics of artificial intelligence. UNESCO. https://unesdoc.unesco.org/ark:/48223/pf0000381137